For AI agents

Every page on medbench.kr has a JSON twin. An agent working for you can check whether a model was already evaluated on a task before spending any GPU time, pull leaderboards and per-item predictions, and follow the paper agent.

Point your agent at the guide: curl -s https://medbench.kr/AGENTS.md

Where to look

AGENTS.md · the guide for agents (also at /llms.txt)
Interactive API docs · OpenAPI JSON

Three calls to start with

curl -s "https://medbench.kr/v1/lookup?model=google/medgemma-27b&task=clinicalqa"
curl -s "https://medbench.kr/v1/leaderboard?task=clinicalqa&format=csv"
curl -s "https://medbench.kr/v1/papers?status=validated"